Example. Populations and samples

Transcription

1 Example Two strains of mice: A and B. Measure cytokine IL10 (in males all the same age) after treatment. A B IL10 A B IL10 (on log scale) Point: We re not interested in these particular mice, but in aspects of the distributions of IL10 values in the two strains. 1 Populations and samples We are interested in the distribution of measurements in the underlying (possibly hypothetical) population. Examples: Infinite number of mice from strain A; cytokine response to treatment. All T cells in a person; respond or not to an antigen. All possible samples from the Baltimore water supply; concentration of cryptospiridium. All possible samples of a particular type of cancer tissue; expression of a certain gene. We can t see the entire population (whether it is real or hypothetical), but we can see a random sample of the population (perhaps a set of independent, replicated measurements). 2

2 Parameters The object of our interest is the population distribution or, in particular, certain numerical attributes of the population distribution (called parameters). Examples: mean median SD proportion = 1 proportion > 40 geometric mean 95th percentile Parameters are usually assigned greek letters (like θ,, and ). 3 Sample data We make n independent measurements (or draw a random sample of size n). This gives X 1, X 2,..., X n independent and identically distributed (iid), following the population distribution. Statistic: Estimator: (estimate) A numerical summary (function) of the X s (that is, of the data). For example, the sample mean, sample SD, etc. A statistic, viewed as estimating some population parameter. We write: ˆθ an estimator of θ ˆp an estimator of p X = ˆ an estimator of S = ˆ an estimator of 4

3 Parameters, estimators, estimates The population mean A parameter A fixed quantity Unknown, but what we want to know X x The sample mean An estimator of A function of the data (the X s) A random quantity The observed sample mean An estimate of A particular realization of the estimator, X A fixed quantity, but the result of a random process. 5 Estimators are random variables Estimators have distributions, means, SDs, etc. X 1, X 2,..., X 10 X

4 Sampling distribution Distribution of X n = 5 Sampling distribution depends on: The type of statistic The population distribution The sample size 0 7 Bias, SE, RMSE Dist'n of sample SD (n=10) Consider ˆθ, an estimator of the parameter θ. Bias: Standard error (SE): RMS error (RMSE): E(ˆθ θ) = E(ˆθ) θ SE(ˆθ) = SD(ˆθ). E{(ˆθ θ) 2 } = (bias) 2 + (SE) 2. 8

5 Example: heights of students 14 Frequency mean = 67.1 SD = Height (inches) 9 Example: heights of students mean = 67.1 SD = 3.9 Dist'n of Sample Ave (n=10) mean = 67.1 SD = Height Height (inches) Dist'n of Sample Ave (n=5) mean = 67.1 SD = 1.7 Dist'n of Sample Ave (n=25) mean = 67.1 SD = Height (inches) Height (inches) 10

6 The sample mean Assume X 1, X 2,..., X n are iid with mean and SD. Mean of X = E( X) =. Bias = E( X) = 0. SE of X = SD( X) = / n. RMS error of X = (bias)2 + (SE) 2 = / n. 11 If the population is normally distributed If X 1, X 2,..., X n are iid normal(,), then X normal(, / n). Distribution of X n 12

7 Example Suppose X 1, X 2,..., X 10 are iid normal(mean=10,sd=4) Then X normal(mean=10, SD 1.26); let Z = ( X 10)/1.26. Pr( X > 12)? % Pr(9.5 < X < 10.5)? 31% Pr( X 10 > 1)? 43% Central limit theorm If X 1, X 2,..., X n are iid with mean and SD. and the sample size (n) is large, then X is approximately normal(, / n). How large is large? It depends on the population distribution. (But, generally, not too large.) 14

8 Example 1 Distribution of X n = Example 2 Distribution of X n =

9 Example 2 (rescaled) Distribution of X n = Example 3 Distribution of X {X i } iid Pr(X i = 0) = 90% Pr(X i = 1) = 10% E(X i ) = 0.1; SD(X i ) = n = 500 X i binomial(n, p) X = proportion of 1 s

10 The sample SD Why use (n 1) in the sample SD? (xi x) 2 s = n 1 If {X i } are iid with mean and SD, then E(s 2 ) = 2 but E{ n 1 n In other words: Bias(s 2 ) = 0 s2 } = n 1 n 2 < 2 but Bias( n 1 n s2 ) = n 1 n 2 2 = 1 n 2 19 The distribution of the sample SD If X 1, X 2,..., X n are iid normal(, ) then the sample SD, s, satisfies (n 1) s 2 / 2 χ 2 n 1 When the X i are not normally distributed, this is not true. χ 2 distributions df=9 df=19 df=

11 Distribution of sample SD (based on normal data) n=25 n=10 n=5 n= A non-normal example Distribution of sample SD n = n =